Building music tech at Beat22. I pick up something new every weekend. Currently failing spectacularly at tennis.
I'm Keshav Garg, a software developer who builds systems that ship and scale.
Currently at Beat22 full-time, where I work on deep audio processing and AI music detection. I built the audio fingerprinting engine from the ground up, an automated internal code reviewer that runs pre-merge checks, and shipped backend systems the team uses every day.
Outside work I'm building an audio production pipeline, a native macOS menu bar utility, a P2P desktop messenger, and a DAW in the browser.
Reach me at me@keshavgarg.com.
Remote
Promoted to full-time after the internship. Continue owning the AI music detection pipeline and the internal code review system, while taking on broader backend ownership and new product features.
Remote
Music is where my interests live, and this role puts me right at that intersection. Built the AI music detection and audio fingerprinting system from the ground up, shipping it to production across a platform with 30,000+ active users. Wrote an internal automated code review system that runs pre-merge checks with no external dependency, improved API response latency across key endpoints, and designed new backend modules the team relies on daily.
Native macOS menu bar app with 40+ system utilities built from scratch in Swift. Clipboard history with search, WiFi network analyzer, port scanner, live CPU and memory graphs, JSON and UUID tools, and a Now Playing widget with full lyrics and Dynamic Island integration. Ships as a single native binary.
Audio production pipeline that takes a raw vocal recording and any instrumental and outputs a properly balanced, mastered stereo mixdown. Aligns tracks by onset detection, applies per-band EQ and dynamic compression matched to genre loudness targets, and exports to broadcast specs. Built to solve the exact problem every bedroom producer hits.
Local AI proxy that eliminates redundant context while preserving byte-level reversibility across all your coding tools. Translates between Anthropic, OpenAI, Gemini, and Ollama wire formats simultaneously. Does context compression, deduplication, and prompt-cache preservation automatically. Ships with model routing, A/B cost measurement, a full CLI, and diagnostics — backed by 874 passing tests.
Desktop messaging over raw TCP with no server in the middle. Both ends run the same binary, discover each other on the local network, and exchange encrypted messages peer-to-peer. Targets Windows, macOS, and Linux from a single codebase. No accounts, no cloud, no data leaving your machine.
A full language implementation built from scratch: lexer, recursive-descent parser, bytecode compiler, and a stack-based virtual machine. Includes a REPL, interactive debugger with breakpoints, assembler and disassembler, and compiler optimizations including constant folding and dead code elimination.
Real-time ride-sharing platform where drivers post available rides and riders book dynamically. PostGIS powers sub-second geospatial search across active rides, with a Turborepo monorepo, Next.js frontend, and PostgreSQL on Supabase keeping the stack tight.
Terminal assistant with Gemini under the hood. Reads and writes files, generates code, debugs errors, translates text, and summarizes documents without leaving the command line. Installable globally from npm.
Real-time exam proctoring that runs entirely on the invigilator's machine. Tracks face presence, gaze direction, head pose, and lip movement simultaneously. YOLO detects phones with a three-strike escalation system. Live Flask dashboard streams session events and alerts.
Best reached via email or LinkedIn. Use the chat bar below to ask anything about my work.
Keyboard shortcuts
